dr inż. Stanisław Gepner,
Wydział Mechaniczny Energetyki i Lotnictwa
Nowowiejska 21/25 Str.
00-665 Warszawa, Polska
tel. +48 (22) 234 51 70
Hi Wang,
KDIM sets the size of the Krylov space used to evaluate updates
to the current eigenvector approximation. In general, the
procedure will run at least the number of iterations equal to
KDIM, each time adding one vector to the sub-space until there is
a KDIM of them (see the .evl file, that stores some information on
the solution process). Than, the procedure start replacing vectors
with new approximations, until convergence criterion is met. My
observation is, that having this value to small can lead to a
large number of iterations that you have to perform, and having it
to large, forces a large number of iteration to start with, but
past KDIM vectors the convergence should be quicker. So I would
advice practicing with a channel test case, to get a feel. In my
opinion this value should have no relation to the final frequency,
but will influence convergence.
NumSteps determines the number of time steps run in the linear
solver. The larger it is, the longer the time that the linear
operator evolves the current perturbation. So IMHO the longer it
is, the more distinct should the leading eigen mode become -
leading to faster convergence. But then, if you have some spurious
osculations in your solution, those would be amplified as well,
obfuscating the convergence.
Finally, I have found, that ModifiedArnoldi driver will in general be rather slowly converging, and I tend to use the Arpack driver, with a relatively large KDIM (64,128 and up). One issue is, Arpack is serial only, so if your case is very large this could be an issue.
This email from wangzhouyang@buaa.edu.cn originates from outside Imperial. Do not click on links and attachments unless you recognise the sender. If you trust the sender, add them to your safe senders list to disable email stamping for this address.
Dear nektar++ users and developers.
Sorry to bother everyone here, however, I need some help the modifiedArnoldi solver.
In the linear stability solver, there are NUMSTEPS and KDIM to be filled in, what is the relationship between the setting of these two variables and the frequency of the unstable mode to be solved?
For example, if the frequency of the unstable mode is approximately 0.1 Hz (0.2pi for the circle frequency) and TIMESTEP = 0.002, should NUMSTEPS and KDIM be set relatively large or small in order to capture this unstable mode?
I wonder if anyone has any relevant experience in setting this up and I would appreciate any help.
Thanks again to the users and developers of nektar++.
Best Regards,
Wang Zhouyang
_______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/nektar-users